Honor /compact summary when auto-summarization is disabled

[bhavyaus]

Fixes #311503.

Problem

With github.copilot.chat.summarizeAgentConversationHistory.enabled: false, the agent renders conversation history via AgentConversationHistory, which replayed every turn's full tool-call history and never looked at round.summary. So after a manual /compact stored a summary, the next turn rebuilt the entire pre-compaction context and the context window meter immediately went red again.

Repro: disable the setting → fill the context → /compact (meter empties) → send another prompt → meter full again.

Fix

The summary data is already available: normalizeSummariesOnRounds() runs on every request (in chatParticipantRequestHandler.ts) and populates round.summary on historical rounds. Only the summarized renderer honored it.

AgentConversationHistory now:

1. Finds the most recent summarized round across history.
2. Drops the turns/rounds it supersedes.
3. Emits the <conversation-summary> tag in place of that turn's user message.
4. Replays only the rounds that follow the summary.

This applies an existing summary but never triggers new summarization — automatic summarization stays gated behind the setting.

Scope / risk

• Single production file changed (agentConversationHistory.tsx); no new props, imports, or exports.
• The common no-summary path is byte-for-byte unchanged (summaryTurnIndex === -1 short-circuits every new branch).
• Distinct from Fix stale background compaction across model switches and /compact #317163, whose logic lives entirely behind summarizationEnabled === true and was therefore inert in this disabled-setting scenario.

Validation

• Typecheck clean (only pre-existing unrelated chatSessions/ errors).
• summarization.spec.tsx: 35/35 pass, including a new regression test for GitHub Copilog github.copilot.chat.summarizeAgentConversationHistory.enabled: false fills context window in single round. #311503.
• agentPrompt.spec.tsx: 165/165 pass.

[Copilot]

Pull request overview

This PR fixes a regression where manual /compact summaries were ignored when github.copilot.chat.summarizeAgentConversationHistory.enabled is disabled, causing the full pre-compaction tool-call history to be replayed and the context meter to immediately refill. It updates the non-summarizing history renderer to honor existing round.summary values (populated by normalizeSummariesOnRounds) by dropping superseded history and replaying only the rounds that follow the summary.

Changes:

• Update AgentConversationHistory to detect the most recent summarized round, omit superseded turns/rounds, and render a <conversation-summary> tag in the prompt.
• Add a regression test ensuring manual /compact summaries are applied even when automatic summarization is disabled.

Show a summary per file

File	Description
extensions/copilot/src/extension/prompts/node/agent/agentConversationHistory.tsx	Honors existing round.summary when rendering history with summarization disabled by dropping superseded history and emitting a <conversation-summary> tag.
extensions/copilot/src/extension/prompts/node/agent/test/summarization.spec.tsx	Adds a regression test covering the “manual /compact summary honored when summarization is disabled” scenario (#311503).

Copilot's findings

• Files reviewed: 2/2 changed files
• Comments generated: 0